Reduce uses of windows.d b/c it's slow (see PR #2336)

[n8sh]

In general avoided removing non-private core.sys.windows.windows imports from other core.sys.windows.* header files so imports-of-imports from them will not break. The exception is core.sys.windows.threadaux because it is imported by core.thread.

[dlang-bot]

Thanks for your pull request, @n8sh!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "master + druntime#2400"

[n8sh]

I can also do this in core.sys.windows.snmp and core.sys.windows.httpext because their imports of core.sys.windows.window are private.

[rainers]

In general avoided removing non-private core.sys.windows.windows imports from other core.sys.windows.* header files so imports-of-imports from them will not break.

import is private by default, so only "public imports" might be affected.

Selective imports do create public symbols, though (albeit deprecated). So this could be a breaking change causing ambiguities before the deprecated behaviour is removed from the compiler. It might be less invasive to not list imported symbols (while mostly desirable in general some lists look a bit excessive).

Do you have some example numbers what the compile time improvements are?

[n8sh]

Selective imports do create public symbols, though (albeit deprecated). So this could be a breaking change causing ambiguities before the deprecated behaviour is removed from the compiler.

Didn't know that, thanks. This PR should not be merged before I fix this.

Do you have some example numbers what the compile time improvements are?

I assumed there would only be a noticeable speed difference with uncached reads from a non-SSD drive (I'm assuming the main speed penalty from importing windows.d is opening ~32 other files), but benchmarking would still be a good idea.

[thewilsonator]

Ping @n8sh

[n8sh]

@thewilsonator removed top-level selective imports as @rainers recommended. I also did an informal benchmark with a non-optimized DMD development build. Compiling the following program went from 640 milliseconds (best of 40 tries) to 240 milliseconds (best of 40 tries):

import core.internal.abort;
import core.runtime;
import core.sync.condition;
import core.sync.mutex;
import core.sync.semaphore;
import core.sys.windows.httpext;
import core.sys.windows.snmp;
import core.sys.windows.threadaux;
import core.thread;
import core.time;

void main(string[] args) {}

[n8sh]

That was a more noticeable difference than I expected. I'll remove the remaining imports of windows.d from other core.sys.windows.* files.